Symbolica launches on $33M to out-design transformers with category theory, not scale

Builds new AI model architectures applying category theory to structured reasoning, instead of scaling transformers

That category theory can yield a whole class of models that outperform transformers — smaller, interpretable, able to actually reason

$2M pre-seed from Khosla in 2023 to prove the idea, then a $33M Series A in April 2024 (valuation undisclosed)

George Morgan spent four years at Tesla as a senior autopilot engineer, starting as an intern, concluded the scale-only philosophy was wrong, and left in 2022 to found Symbolica

Khosla gave $2M in 2023 — 'He delivered that, very credibly' — then told Morgan to hire the best category theorists. Symbolica published its approach with Google DeepMind researchers and launched April 9, 2024 with the $33M round

Every major lab was scaling transformers with training runs costing hundreds of millions; Symbolica's bet is that architecture, not scale, is the next lever.

The founder's conviction was forged at Tesla, where more scale kept failing to crack full self-driving — the highest-profile stress test of scale-only thinking.

Khosla turned it into a staged bet: $2 million to prove the math first, $33 million only after Morgan 'delivered that, very credibly.'

When a field converges on one recipe, the highest-leverage bet is the credible outsider who thinks the recipe itself is the bottleneck.

As of the April 2024 launch, Symbolica said it was building a class of models rather than a single transformer-beating model; the Series A funds hiring category theorists and advancing the architecture. Valuation was not disclosed.